Folia Neuropathologica
eISSN: 1509-572x
ISSN: 1641-4640
Folia Neuropathologica
Current issue Archive Manuscripts accepted About the journal Special Issues Editorial board Reviewers Abstracting and indexing Subscription Contact Instructions for authors Publication charge Ethical standards and procedures
Editorial System
Submit your Manuscript
SCImago Journal & Country Rank
Search
Editorial Policies
Sarajevo Declaration on Integrity and Visibility of Scholarly Publications
3/2025
vol. 63
 
Share:
Send email
Share:
abstract:
Original paper

Application of a convolutional neural network model to construct an automatic, AI-based identification system for rat kidney tissue microscopic images

Chen Yun
1, 2
,
Ai-Yun Zha
2, 3
,
Xiaofan He
4
,
Suhua Xu
4
,
Xiaohong Tang
5
,
Qiang Li
6
,
Lianghong Yin
2, 3
,
Shaodong Luan
7

  1. Department of Nephrology, Charité – Universitätsmedizin Berlin, Campus Mitte, Berlin, Germany
  2. Clinical Medical Research Institute of the First Affiliated Hospital of Jinan University, Guangzhou 510632, Guangdong Province, China
  3. Department of Nephrology, The First Affiliated Hospital of Jinan University, Jinan University, Guangzhou 510632, Guangdong Province, China
  4. Guangdong Medical Devices Quality Surveillance and Test Institute, Guangzhou 510663, Guangdong Province, China
  5. Department of Nephrology, The Second Affiliated Hospital of Guangzhou Medical University, 510260 China
  6. Department of Nephrology, Dongguan Hospital of Traditional Chinese Medicine, Dongguan 523000, China
  7. Department of Nephrology, Shenzhen Longhua District Central Hospital, Shenzhen 518110, Guangdong Province, China
Folia Neuropathol 2025; 63 (3): 289-303
DOI: https://doi.org/10.5114/fn.2025.153859
Online publish date: 2025/09/16
View full text Get citation
 
PlumX metrics:
The aim was to develop an artificial intelligence (AI)-based identification system using a deep learning method based on convolutional neural networks to analyze pathological kidney images. We constructed a rat chronic kidney disease (CKD) model and used different drug interventions to study the pathological changes in the kidneys, which is convenient for machine in-depth learning and the construction of a recognition system. The reliability and credibility of the system were assessed by a blind comparative analysis. Microscopic image recognition and classification: Five pathological groups were subjected to three types of staining to obtain 15 image groups. Fifty jpg images were captured from each image group, so that 750 images were captured for training. The average hard disk space per image was 354 kb. Because the input image was 1000 pixels wide with a large resolution, multiple sampling was required to extract feature information. This implies that the level of span multiplication greatly affected the performance of the network during sampling. First, the ResPool samples with the same number of channels, followed by the output of a convolution layer, were assembled as the incremental channel. Thus, the feature extraction network of this work successfully implemented the idea of residual learning and did not introduce errors in the deep network to achieve the expected effect. The neural network model of this study designed a ResPool sampling structure based on the idea of residual learning, completed glomerular instance segmentation and damage analysis, and improved the accuracy of image recognition tasks.
keywords:

convolutional neural network model, CKD model, CKD pathological change map, automatic identification of microscopic images, residual learning

  
Termedia
About us Contact
Copyright notice Privacy policy Advertising policy Contact us
Quick links
Journals Books eBooks Events
© 2025 Termedia Sp. z o.o.
Developed by Bentus.